Simulations-Of-Snow-Management.Pdf

Cold Regions Science and Technology 172 (2020) 102995

Contents lists available at ScienceDirect

Cold Regions Science and Technology

journal homepage: www.elsevier.com/locate/coldregions

Simulation of snow management in Alpine ski resorts using three diﬀerent snow models T ⁎ Florian Hanzera,b, , Carlo Maria Carmagnolac, Pirmin Philipp Ebnerd, Franziska Koche, Fabiano Montif, Mathias Bavayd, Matthias Bernhardte, Matthieu Lafayssec, Michael Lehningd,g, Ulrich Strassera, Hugues Françoish, Samuel Morinc a Department of Geography, University of Innsbruck, Austria b Wegener Center for Climate and Global Change, University of Graz, Austria c Univ. Grenoble Alpes, Université de Toulouse, Météo-France, CNRS, CNRM, Centre d’Etudes de la Neige, Grenoble, France d WSL Institute for Snow and Avalanche Research SLF, Davos, Switzerland e Institute of Hydrology and Water Management, University of Natural Resources and Life Sciences (BOKU), Vienna, Austria f ALPsolut S. r. l., Livigno, Italy g School of Architecture, Civil and Environmental Engineering, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland h Univ. Grenoble Alpes, Irstea, LESSEM, Grenoble, France

ARTICLE INFO ABSTRACT

Keywords: Snow management, i. e., snowmaking and grooming, is an integral part of modern ski resort management. While Snow management the current snow cover distribution on the slopes is often well known thanks to the usage of advanced mon- Ski resorts itoring techniques, information about its future evolution is usually lacking. Management-enabled numerical Snowpack modeling snowpack models driven by meteorological forecasts can help to fill this gap. In the frame of the H2020 project PROSNOW, the snowpack models AMUNDSEN, Crocus, and SNOWPACK/ Alpine3D are applied in nine pilot ski resorts across the European Alps for forecasting snow conditions in time scales from days to several months ahead. We present the integration of detailed snowmaking and grooming practices implemented in the three models and show how they can be adapted to individual ski resorts. An ensemble of snow management configurations accounting for a comprehensive set of possible tactical and strategic operational choices is introduced, along with an approach to homogeneously spatialize the results of the three snow models over different areas of the ski resorts. First simulation results are presented for the nine pilot ski resorts in the form of distributed snow water equivalent (SWE) maps along with SWE and snow depth time series for two selected seasons in the past.

1. Introduction aim to optimize their snowmaking practices in order to produce snow most efficiently, i. e., in terms of timing, volume, and costs. Often, ski During the past decades, ski resorts throughout the world have resort managers have to make important decisions under anticipation of become increasingly reliant on snowmaking facilities to complement future weather and snow conditions. This includes, e. g., identifying the the natural snow cover. Primarily this is motivated by the desire for on- trade-off between producing snow during marginally cold periods vs. time planning of ski resort operations and efficient managing of ap- waiting for a colder period potentially coming up in the next few weeks propriate skiing conditions on the slopes independently of usually (while risking ending up with not enough snow on the slopes), or de- highly variable natural snowfall amounts. The goal is to have well termining the optimal start date of the base-layer snow production to controllable snow conditions at the right time, which are smooth and ensure adequate preparation of the slopes in time for the planned even and robust enough to withstand the impact of today’s usual opening date (while risking for a warm spell to melt all of the produced number of skiers (Spandre et al., 2015; Hanzer et al., 2014). As snow- snow again). Although it is increasingly common to employ advanced making operations require considerable investments both in terms of monitoring techniques such as GPS-equipped grooming devices infrastructure and the use of resources (water and energy), ski resorts tracking the snow depth on the slopes, most ski resort managers rely

⁎ Corresponding author at: Department of Geography, University of Innsbruck, Austria. E-mail address: fl[email protected] (F. Hanzer). https://doi.org/10.1016/j.coldregions.2020.102995 Received 29 March 2019; Received in revised form 6 December 2019; Accepted 15 January 2020 Available online 22 January 2020 0165-232X/ © 2020 Published by Elsevier B.V. F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995 solely on short-term weather forecasts for planning snowmaking op- investigated. Results of snow simulations in terms of snow depth and erations, while information about the future evolution of the snowpack snow water equivalent (SWE, the mass of snow per surface unit) have is usually lacking. Given the appropriate initial conditions and me- been obtained driving the snow models with reanalyzed forcing data- teorological forecasts, this gap can be filled by applying numerical sets. A first comparison of the model outputs with natural snow depth snowpack models accounting for snow management practices, i. e., the values measured by in-situ weather stations is presented. physical descriptions of the snowmaking and grooming processes and The paper is structured as follows. After a short general introduction the associated socioeconomic decisions. to the three models, we present an overview and comparison of the Several simple snow models integrating rule-based snow manage- current state of integration of snow management and the available ment have been developed during the past decades (see Steiger and parameters allowing to adapt the models to resort-specific management Abegg (2013) for references), mainly in order to assess the impact of practices. Subsequently, we describe the operational workflow of cal- climate change on snow reliability and ski tourism activity. In these culating the snowpack forecasts within PROSNOW in terms of the approaches, the snowmaking strategies are based on general assump- spatial clustering of ski resorts and the management configurations tions developed in stakeholder work for selected case study destina- used for the simulations, as well as the setup of the models for the tions. However, the physical properties of snow were represented in a individual ski resorts. Finally, simulation results are presented for all comparatively coarse manner, and snowmaking (or snowmaking po- nine pilot ski resorts in the form of time series of snow depth and SWE tential) was accounted for using a few simple assumptions. The ex- evolution and distributed SWE maps for two selected seasons of the past planatory power of scenario simulations with these simple models is and three different snow management configurations. hence limited to estimates of regional patterns of potential climate change impacts (Abegg et al., 2020). This inspired the development of a new generation of management-enabled snowpack models during the 2. Study sites and data past few years, most notably by the studies by Hanzer et al. (2014) and Spandre et al. (2016b) who integrated snow management into the 2.1. Pilot ski resorts physically based snowpack models AMUNDSEN (Strasser, 2008) and Crocus (Vionnet et al., 2012), respectively. Recently, snow management The nine PROSNOW pilot ski resorts that are part of this study are: practices have also been integrated into the SNOWPACK/Alpine3D Seefeld (cross-country part) and Obergurgl in Austria, Colfosco, San model (Bartelt and Lehning, 2002; Lehning et al., 2006). As these three Vigilio, and Livigno in Italy, La Plagne and Les Saisies in France, Arosa- snowpack models have different backgrounds and original purposes (e. Lenzerheide in Switzerland, and Garmisch Classic/Zugspitze in g., hydrology vs. avalanche forecasting (Morin et al., 2019)) they differ Germany. The choice of these resorts allows representing a large di- in their design and internals (e. g., in terms of onedimensional vs. versity of geographical, climatical and technical characteristics, infra- spatially distributed application, representation of snowpack layering structure and snowmaking equipment, as well as governance setting and microstructure, regionalization of meteorological variables), how- and economic dynamics. ever they have in common that they are capable of representing the ski Snowpack simulations are performed with AMUNDSEN for the resort infrastructure and the specifics of the snowmaking and grooming Austrian and South Tyrolean (Colfosco and San Vigilio) resorts, with processes in a considerable level of detail. As it has been demonstrated Crocus for the French resorts, and with SNOWPACK/Alpine3D for the that they are able to adequately reproduce the real snow conditions on remaining resorts in Switzerland, Germany, and Italy (Arosa- the slopes, they are not only applicable for long-term climate change Lenzerheide, Garmisch Classic/Zugspitze, and Livigno). Fig. 1 and impact studies (Marke et al., 2015; Spandre et al., 2018) but also for Table 1 show the locations and key characteristics of the pilot resorts. potential real-time applications. These three models are currently being applied in the frame of the H2020 PROSNOW project (Morin et al., 2018), where a demonstrator of 2.2. Model input data a meteorological and snow prediction system for time scales ranging between several days and several months ahead is being developed. For For all three models, required spatial input data for the snow nine ski resorts across the Alps, state-of-the-art meteorological and management simulations consists of a digital elevation model (DEM) climate forecasts will be used to feed AMUNDSEN, Crocus, and SNO- covering the study sites, the locations of the ski slopes, and the locations WPACK/Alpine3D and deliver information of the future snowpack and types (at the minimum divided into lances and fans) of the snow evolution depending on both the meteorological forecasts and possible guns. Meteorological forcing data for the simulations is based on snow management approaches. Ultimately, these forecasts are intended measurements from automatic weather stations in proximity to the to help ski resort managers in anticipating important decisions such as study sites, consisting of at least hourly measurements of air tempera- inhibiting production under conditions where the produced snow might ture, precipitation, relative humidity, wind speed, and radiation. be subject to melting, or identifying the periods which correspond to the most favorable snowmaking conditions while maintaining reliable snow conditions on the slopes. In this article, we present the integration of the snow management practices into the three models, along with the first results of the simulations carried over the PROSNOW pilot ski resorts. The goal of this work is twofold. On the one hand, we introduce here an ensemble of snow management configurations accounting for a comprehensive set of possible tactical and strategic operational choices. Those configurations define the common framework used by the different snowpack models to represent consistently the eff ects of snow management on the snow physical properties, while accounting for the specific practices of individual ski resorts. On the other hand, this work also presents an approach to spatialize the results of the modeling chain over different areas of the ski resorts. With the aim of spatially homogenizing the results across snowpack models, a trade-off between the geometry of the model outputs and the operational needs of the ski resorts has been Fig. 1. Locations of the PROSNOW pilot ski resorts.

2 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Table 1 we describe the general workﬂow and implementation of snowmaking Overview of the nine PROSNOW pilot ski resorts. in the three snowpack models and how these factors are accounted for

Resort Country Elevation range (m Slope surface area by discussing the snow production related parameters in the models. a.s.l.) (ha) These parameters, listed in Table 3, allow to adjust the simulation of snow production according to the infrastructure and snowmaking – Arosa-Lenzerheide CH 1200 2865 384 practices of individual ski resorts. A ﬂowchart of the core snow pro- Colfosco IT 1531–2218 64 Garmisch Classic/ DE 708–2720 66 duction procedure as described below is shown in Fig. 2. Zugspitze La Plagne FR 1250–3250 528 Les Saisies FR 1150–2069 214 3.1.1. Snow demand Livigno IT 1816–2797 448 The calculation of (i), snow demand, i. e., when and where in the ski Obergurgl AT 1930–2898 107 resort snow should be produced, is determined by the production San Vigilio IT 1087–2274 119 period (PP) and production time (PT) parameters, which are both – Seefeld AT 1179 1251 79 speciﬁed separately for the base-layer and the reinforcement period

(PPb,PPr,PTb,PTr), as well as the consumption threshold (CT) and 3. Models snow threshold (STb,STr) parameters. These parameters are described below in detail. fi All three employed snowpack models, AMUNDSEN, Crocus, and PPb and PPr de ne the base-layer and reinforcement snowmaking SNOWPACK/Alpine3D, are well-established and have been widely ap- periods (in the form of a start and end date), i. e., the periods where plied in numerous studies throughout the past decades. While an snowmaking is generally possible (which are in practice determined overview of the most important model specifics including a list of re- both by legal regulations and by economical and practical considera- ferences for further information is shown in Table 2, in this article we tions such as the planned season opening date). Similarly, PTb and PTr fi refrain from a more general model description and focus in the fol- de ne the periods (in the form of a start and end time) within each day lowing only on the description of the integration of snow management of the production period in which snowmaking should be performed (e. processes. For details on the model internals, the distribution of me- g., during the base-layer period snow could be produced all day, teorological variables, and the calculation of natural snow processes we whereas during the reinforcement period snow should only be produced refer to the respective literature references. during nighttime when no skiers are on the slopes and temperatures are In the following two sections, the functionality and parameters of generally lower). – – the snowmaking and grooming modules of the snowpack models are In addition to the date and time, the decision if and how much fl described. For AMUNDSEN and Crocus, the described functionality snow should be produced can also be in uenced by the amount of snow corresponds for the most part to the more detailed descriptions pre- already produced and by the current snow conditions on the slopes. sented in the respective studies by Hanzer et al. (2014) and Spandre Common practices are to aim at a certain target snow depth or pro- et al. (2016b), however also includes several refinements and adapta- duction volume (depending on the available water resources) during tions that have since been incorporated. The snow management module the base-layer period, while during the reinforcement period snow is of SNOWPACK/Alpine3D has been implemented from scratch through only produced when snow depth on the slopes is below a critical the PROSNOW project and is herein presented for the first time. threshold (e.g., Spandre et al., 2016a; Steiger et al., 2017). In the models this is accounted for by the consumption threshold (CT; kg − m 2) and the snow threshold (ST ,ST; m) parameters. CT (im- 3.1. Snow production b r plemented in AMUNDSEN and Crocus) allows to specify a certain target water consumption volume which should be met before production is The production of snow in ski resorts is influenced by several fac- stopped during the base-layer period. The snow threshold parameter tors, most importantly (i) snow demand (i. e., if there is a need for (implemented as ST for the base-layer period (only available in SNO- producing snow at a given location within the resort), (ii) adequate b WPACK/Alpine3D) and as ST for the reinforcement period (available in ambient conditions (cold and dry enough air allowing to produce snow, r all three models)) on the other hand allows to specify that snow should low wind speeds for avoiding blowing snow losses), and (iii) ski resort be produced until the snow depth on the slopes (i. e., the sum of natural infrastructure and availability of resources (e. g., number and efficiency and machine made snow) exceeds a certain value. of snow guns, water availability, pumping capacity). In practice, the ski season is commonly divided into a base-layer snowmaking period prior to the opening of the resort where snow is produced whenever possible 3.1.2. Ambient conditions depending on the ambient conditions and a reinforcement snowmaking With regard to ambient conditions (ii), both a wet-bulb temperature − period afterwards, where snow is produced more selectively depending threshold (TT; °C) and a wind speed threshold (WT; m s 1) can be set in on demand and often only during times when no skiers are on the slopes the models in order to account for conditions where snowmaking is (e.g., Spandre et al., 2016a; Steiger and Mayer, 2008). In the following inefficient (marginal temperatures, too high wind speeds) or not

Table 2 Overview of the structure and input data of the AMUNDSEN, Crocus, and SNOWPACK/Alpine3D snowpack models. Meteorological variables are: air temperature (T), total/solid/liquid precipitation (P/Ps/Pr), relative humidity (RH), shortwave/longwave radiation (Rs/Rl), and wind speed (WS).

AMUNDSEN Crocus SNOWPACK/Alpine3D

Key reference(s) Strasser (2008); Strasser et al. (2011); Hanzer Vionnet et al. (2012); Lafaysse et al. Bartelt and Lehning (2002); Lehning et al. et al. (2016) (2017) (2006) Spatial scale Distributed Point scale Point scale (SNOWPACK)/Distributed (Alpine3D) Vertical snowpack discretization 2–4 bulk layers Multi-layer Multi-layer

Meteorological input data T, P, RH, Rs,WS T,Ps,Pr, RH, Rs,Rl, WS T, P, RH, Rs,Rl,WS Temporal resolution of input data 1–3 h 1 h 30min–24h Meteorological preprocessing Built-in Often associated with SAFRAN (Durand MeteoIO (Bavay and Egger, 2014) et al., 1993)

3 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Table 3 Adjustable snow production related parameters as implemented in the three models. The last three columns indicate whether the respective parameter is implemented in AMUNDSEN (A), Crocus (C), or SNOWPACK/Alpine3D (S).

Parameter Symbol Unit Function of Description ACS

Snow demand

Base-layer production period PPb Date range Start and end date for base-layer snowmaking × × ×

Reinforcement production period PPr Date range Start and end date for reinforcement snowmaking × × ×

Base-layer production time PTb Time range Daily start and end time for snowmaking during the base-layer period × × ×

Reinforcement production time PTr Time range Daily start and end time for snowmaking during the reinforcement period × × × − Consumption threshold CT kg m 2 Water consumption threshold (SWE equivalent) for stopping production during the ××– base-layer period

Base-layer snow threshold STb m Snow depth threshold for stopping production during the base-layer period ––×

Reinforcement snow threshold STr m Snow depth threshold for stopping production during the reinforcement period × × × Ambient conditions Temperature threshold TT °C Snow gun type Wet-bulb temperature threshold for snowmaking × × × − Wind threshold WT m s 1 Wind speed threshold for snowmaking × × ×

Ski resort infrastructure and available resources Number of snow guns NG – Slope Number of snow guns for each ski slope × – × Snow spreading surface SS m2 Surface area covered by a snow gun – × – 3 −1 −1 Production rate parameter 1 PRa m h ∘ C Snow gun type First parameter of Eq. (1) to calculate the water flow rate for a single snow gun × × × 3 −1 Production rate parameter 2 PRb m h Snow gun type Second parameter of Eq. (1) to calculate the water flow rate for a single snow gun × × × Water availability WA m3 Total water volume available for snowmaking × – × − Refill rate RR m3h 1 Water refill rate × – × − Water flow threshold FT m3h 1 Maximum total water flow × – ×

Snow properties Water losses WL – Fraction of water lost due to thermodynamic and mechanical eﬀects × × × −3 a Density ρmm kg m Density of machine-made snow × × – 2 −1 SSA SSAmm m kg Speciﬁc surface area of machine-made snow – ××

Sphericity Smm % Sphericity of machine-made snow – ××

a Density is parameterized according to Eq. (2)

Current date and time No Current date and time within PPb and PT b? within PPr and PT r? Yes Yes (AMUNDSEN / Crocus) (SNOWPACK/Alpine3D) Yes

Yes Cum. production Snow depth < ST ? Snow depth < ST ? < CT? b r

Yes

Tw < TT?

Yes

Wind speed < WT?

Yes

WA > 0? (AMUNDSEN / (AMUNDSEN

SNOWPACK/Alpine3D) Yes

Calculate PRReduce accor- Reduce accor- Distribute snow Update WA ding to FT ding to WL

(AMUNDSEN / SNOWPACK/Alpine3D) (AMUNDSEN / SNOWPACK/Alpine3D)

Fig. 2. Flowchart of the snow production procedure in the three models as described in Section 3.1 and Table 3. For all cases not explicitly covered in the chart, no snow is produced.

4 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Table 4 Adjustable grooming related parameters as implemented in the three models. The last three columns indicate whether the respective parameter is implemented in AMUNDSEN (A), Crocus (C), or SNOWPACK/Alpine3D (S).

Parameter Symbol Unit Description ACS

Grooming period GP Date range Start and end date for grooming ××× Grooming time GT Time range Daily start and end time for grooming ××× Grooming threshold GH kg m−2 or m Minimum SWE (AMUNDSEN/Crocus) or snow depth (SNOWPACK/Alpine3D) required for grooming × × × − Penetration depth PD kg m 2 or m Part of the snowpack affected by grooming (specified as SWE in Crocus and as snow depth in SNOWPACK/Alpine3D) – ×× −3 Target density ρt kg m Target density that could be reached by grooming × × × 2 −1 Target SSA SSAt m kg Target specific surface area that could be reached by grooming – × –

Target sphericity St % Target sphericity that could be reached by grooming – × – possible at all. If at least one of these thresholds is exceeded for a given as defined by the WL parameter. −3 location, no snow is produced there. The density of freshly produced technical snow, ρmm (kg m ), is a fixed parameter value in AMUNDSEN and Crocus, whereas in SNOW- 3.1.3. Ski resort infrastructure and availability of resources PACK/Alpine3D it is modeled as a function of the wet-bulb temperature Finally, if all conditions according to (i) and (ii) are fulfilled, the Tw (°C): amount of snow that is actually produced is determined according to 2 ρ =++1.7261TT 37.484w 605.05. (2) (iii), the ski resort infrastructure and available resources. mm w 3 −1 −1 3 −1 The parameters PRa (m h °C ) and PRb (m h ) are used for − calculating the production rate PR (m3 h 1), i. e., the amount of water that can be converted into snow for a given snow gun and time step. PR 3.2. Grooming is assumed to be a linear function of the wet-bulb temperature Tw (°C) at the snow gun location: While in practice grooming in ski resorts is performed both in order to redistribute and to compact snow on the slopes, the former is not PRPRPR=+awT b (1) accounted for by the models, i. e., no explicit transport of snow on the Total snow production volumes for the entire ski resort are calcu- slopes due to grooming takes place. The underlying assumptions are lated differently in the three models. In AMUNDSEN and SNOWPACK/ that freshly produced technical snow is in the model simulations im- Alpine3D, being set up as fully spatially distributed models, for each mediately distributed evenly over the respective slope surface area, and slope (or slope segment) the number of snow guns (NG) to be placed that the entire snow that is transported downwards due to skiers and needs to be supplied. Each slope is then divided into segments of equal wind is later moved back to its original location by the groomers at least area according to the number of snow guns. The model snow guns are daily. The effects of grooming on snow properties (most importantly placed in the center of each segment and produce snow according to the densification) are however explicitly accounted for. Several parameters, meteorological conditions at these specific locations. In Crocus on the listed in Table 4, allow to adjust the schedule and impacts of grooming other hand, the snowmaking module is designed to be applied at the in the individual models as described below. point scale: snow production volumes are calculated according to the Similar to the simulation of snow production, the period and timing meteorological conditions at the simulation point and then scaled ac- of grooming is controlled by the grooming period (GP) and grooming cording to the snow spreading surface parameter (SS), which defines time (GT) parameters. Moreover, as a certain minimum amount of snow the surface area covered by a snow gun (i. e., for a given ski resort, is required, grooming is only performed when and where SWE (AMU- slope, or slope section, this parameter corresponds to the respective NDSEN and Crocus) or snow depth (SNOWPACK/Alpine3D) is above a surface area divided by the number of snow guns located there). specified threshold (GH). In AMUNDSEN and SNOWPACK/Alpine3D, additional snowmaking In Crocus, the densification of the snowpack due to the weight of the infrastructure specifics can be incorporated using the water availability groomer is calculated by applying a static stress of 5 kPa to the topmost − − (WA), refill rate (RR), and water flow threshold (FT) parameters. WA 50 kg m 2 of snow, then linearly decreasing to 0 kPa at 150 kg m 2 of (m3)defines the total water availability for snowmaking at the start of snow. Additional effects due to the tiller mounted to the groomer are the season, which is then reduced during the season according to the applied to the parts of the snowpack specified by the PD parameter (the − − snow guns’ water consumption and increased according to RR (m3 h 1) topmost 35 kg m 2 by default): densification is parameterized as in each time step. If all water resources are depleted, snow production is 3 −1 23ρρ+ stopped. The water flow threshold (FT; m h ) parameter on the other ⎧ av t ⎫ ρgroomed= maxρ av , , hand allows to specify that the total water throughput for the entire ⎩⎨ 5 ⎭⎬ (3) resort is limited (as determined by the pumping and piping infra- where ρ is the weighted average density of impacted layers before structure) – if simulated potential production rates exceed this value, av grooming and ρ is the target density that should eventually be reached production for each snow gun is limited accordingly. In Crocus, these t by grooming (Spandre et al., 2016b). The specific surface area (SSA) parameters are currently not implemented. and sphericity (S) of snow are altered analogously using the respective

target values St and SSAt. Similarly, in AMUNDSEN the bulk snowpack 3.1.4. Water losses and snow properties density of snow on the slopes is altered according to Eq. (3). In SNO- In practice, parts of the water volumes exiting the snow guns ac- WPACK/Alpine3D, similarly to Crocus the PD parameter specifies the cording to Eq. (1) do not reach the ground of the ski slopes in the form part of the snowpack affected by grooming, however here referring to a of snow due to both thermodynamic (evaporation and sublimation) and snow depth (0.4 m by default) rather than snow mass. Densification of ff mechanical (wind-driven redistribution) e ects. While these water the affected snowpack layers during a single grooming run is calculated fi losses can be signi cant even under ideal conditions (Spandre et al., as 2017; Grünewald and Wolfsperger, 2019), simulating them using phy- 0.5 sical formulations is challenging except for simple estimations of the ρgroomed =−+−12.152(448.78ρρ ) 0.9963 35.41 (4) losses due to thermodynamic effects such as applied in Hanzer et al. − (2014). Rather, all three models allow to assume a fixed water loss ratio for ρ ≤ 450kgm 3.

5 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Fig. 4. Aggregation of gridded snow model output data (left, 10 m resolution) for a ski slope to the SRU scale using two exemplary SRU discretizations (center and right).

data or snow depth measurements, will be processed on the SRU scale.

4.2. Model conﬁgurations

While the formulation of snow management processes in the models generally allows to adequately simulate operational practices (as demonstrated previously by Hanzer et al. (2014) for an Austrian ski resort and Spandre et al. (2016b, 2017) for French ski resorts), in practice the Fig. 3. Example of the discretization of a ski resort (Les Saisies) into SRUs based parameters listed in Table 3 are not constant but vary both in space and on the topographic characteristics of the slopes and the presence/absence of time, as the decision when and where to produce snow is made on a snow guns. SRU colors were chosen arbitrarily for visualization purposes. day-by-day basis by the snow production teams in the ski resorts. In order to be able to assist in making these decisions, the operational 4. Model setup and workflow forecasting system developed in PROSNOW is designed to include configurations of different snow management approaches based on the 4.1. Spatial clustering current snow conditions and the meteorological forecasts. As allowing to change the model parameters interactively would not be feasible As in PROSNOW three different models are applied for a range of ski both due to the computational demands and the increasing complexity resorts operationally, it is necessary to agree on a common under- of such a system, a predefined set of configurations which should be standing of the way ski resorts are represented geographically, both for representative of the most important management choices will be technical reasons and for providing information to be used by snow prepared for each ski resort. managers in operational forecast mode of the simulations in a homo- In this work, a single set of parameterizations, according to Table 5, geneous way. In the jointly developed approach, each ski resort is divided into a number of so-called ski resort reference units (SRUs), si- Table 5 milar to the hydrological response units (HRUs) commonly used in Default parameter values (to be adapted for individual ski resorts) for the snow fi hydrological modeling (e.g., Flügel, 1995). management con gurations as described in Section 4.2. Numbered items are ff SRUs are defined by their location and geometry as well as other used to indicate di erent combinations of parameter values. Parameters are according to the definitions in Section 3.1 and Table 3 (IS refers to the in- metadata such as if they are covered by snowmaking facilities or if they hibition switch as introduced in Section 4.2). are being groomed. The concrete delineation of SRUs for a given ski resort can be based on characteristics such as terrain elevation, slope, Parameter Value Combinations aspect, if they are located on a ski slope, the presence or absence of PPb PPb = 01 Nov–15 Dec 1 snow guns, or production priorities. It is estimated that the total PPr PPr = 16 Dec–31 Mar 1 number of SRUs for an average ski resort will typically range between PTb PTb = 00:00–24:00 1 – several tens and a few hundreds. Fig. 3 exemplarily shows a possible PTr PTr = 18:00 08:00 1 3 −1 −1 PRa,PRb 1. Fan guns: PRa = − 1.93 m h ∘ C , discretization of an entire ski resort into SRUs. 3 −1 PRb = 1.58m h While the SRUs constitute the elementary objects that are processed 3 −1 −1 2. Lance guns: PRa = − 1.58 m h ∘ C , 2 3 −1 by the PROSNOW platform and the basis on which model outputs are PRb = − 1.69m h − ∘ delivered, they are not necessarily equivalent to the smallest model TT 1. TTfan = 2 C − ∘ units for which the simulations are performed. Rather, simulations can TTlance = 4 C2 2. TT = − 4 ∘ C still be performed fully spatially distributed on a high resolution grid fan TTlance = − 6 ∘ C − and only be aggregated to the coarser SRU scale in a post-processing WT WT = 4.2ms 1 1 − step. This approach, as illustrated in Fig. 4 using two exemplary SRU CT 1. CT = 150kgm 2 − discretizations for a ski slope, is pursued in the AMUNDSEN and 2. CT = 250kgm 2 2 SNOWPACK/Alpine3D simulations, while the Crocus simulations are STr STr = 0.6m 1 IS 1. IS = (0,0) directly performed on the actual SRU scale. Similarly, local measure- 2. IS = (1,0) 4 ments from the ski resorts that will be assimilated into the snow model 3. IS = (0,1) simulations for the operational model runs, such as water consumption 4. IS = (1,0)

6 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995 has been chosen to configure the snowpack models. This makes it easier by dividing the slopes using an automatic approach taking into account to present and compare the results of the simulations. In the PROSNOW their geographic properties (elevation, aspect, and slope) and the pre- modeling chain, however, all values of Table 5 will be adapted to match sence or absence of snowmaking, resulting in 211 and 374 SRUs for Les the specific needs of the ski resorts. In practice, each ski resort will have Saisies and La Plagne, respectively, with an average area of 1.3 ha. its own set of parameterizations and the possibility to change it every For AMUNDSEN, gridded hourly meteorological input data was year before the beginning of the winter season. The set of configura- provided by AMUNDSEN’s internal meteorological preprocessor using tions according to Table 5 contains a range of strategic variables as observations from stations surrounding the ski resorts. Interpolated presented in Table 3 as well as one additional tactical variable, the so- fields from the point measurements are – in the case of temperature, called inhibition switch (IS). The strategic variables concern the snow precipitation, humidity, and wind speed – obtained using a combined management choices over the entire season, whereas IS allows to define lapse rate–inverse distance weighting scheme, either using auto- a set of rules that guide the daily operational choices in the next few matically calculated lapse rates for each time step or using prescribed days. The default set of parameterizations can be summarized as fol- monthly lapse rates. Radiation fluxes are calculated for clear-sky con- lows: ditions using the parameterizations from Corripio (2002) and corrected

• The base-layer production period (PPb) is set to the period from 1 using cloudiness ﬁelds (Hanzer et al., 2016). Similar approaches were November to 15 December in order for the resorts to be able to open in applied for the SNOWPACK/Alpine3D simulations by using the me- time for the Christmas holidays. During this period production is pos- teorological preprocessing library MeteoIO (Bavay and Egger, 2014). sible during the entire day (PTb = 00:00–24:00). Simulations are per- For Crocus, the generation of consistent meteorological input data was formed for two consumption thresholds (CT) after which production carried out by the meteorological downscaling and surface analysis tool − should be stopped, namely 150 and 250 kg m 2, respectively (corre- SAFRAN (Durand et al., 1993). SAFRAN operates at the geographical sponding to snow depths of 30 and 50 cm assuming an average density scale of meteorologically homogeneous mountain ranges (so-called − of 500 kg m 3). No snow threshold (ST) is set, i. e., these snow amounts “massifs”), within which meteorological conditions are assumed to are produced regardless of the natural snow accumulation during this depend only on altitude and aspect. For the analysis of meteorological period. surface ﬁelds, the guess used by SAFRAN consists of vertical atmo-

• The reinforcement snowmaking period (PPr) is set to the period spheric profiles from numerical models. A robust assimilation scheme 16 December until 31 March. Here, snow is only produced during corrects the initial guess based on ground-based and radiosonde ob- nighttime (PTr = 18:00–08:00) and only if the total snow depth on the servations as well as remotely-sensed cloudiness. Thus, SAFRAN pro- slopes is below STr = 0.6 m (Hanzer et al., 2014). vides hourly meteorological conditions for each massif for 300 m- • Separate simulations are performed assuming the ski resort being spaced elevation bands. equipped with fan guns and lance guns, respectively, using the generic For setting up the snow management modules, the 34 standard parameterizations described in Hanzer et al. (2014). For each of these configurations according to Fig. 5 were used. Model parameters not snow gun types (corresponding to different production rates for given covered by these parameter sets (e. g., the resort-specific infrastructure- ambient conditions) again two scenarios are considered when produc- related parameters) were adapted to each ski resort using available data tion should be triggered: for fan guns, wet-bulb temperature thresholds provided by the ski resorts where possible, or set to reasonable default (TT) of −3 and −4 ∘ C are considered, while for lance guns the values otherwise. It should be remarked that for some resorts water thresholds are set to −4 and −6 ∘ C. These thresholds are aimed to availability (WA parameter) was set to be limited (e. g., to the total represent a comparatively aggressive (i. e., producing snow whenever reservoir volume), while for other resorts it was assumed to be un- possible even in only marginally cold conditions) and a more con- limited either because no exact values were available or because the servative snowmaking strategy, respectively. WA parameter is not implemented in the respective snowpack model (i. • The inhibition switch (IS) aims at accounting for the short-term e., for the French resorts). However, the limited water availability is management decisions by allowing to stop snow production on a daily already implicitly included in the base-layer production targets (CT and basis within the next two days. IS is defined as a tuple (ISd+1,ISd+2), STb parameters), hence explicitly setting the WA parameter will in most where a value of 1 for ISd+1 or ISd+2 indicates that production should cases not change the simulated production volumes, since reinforce- be stopped from 18:00 today (in terms of the current model time step) ment production is usually very minor compared to the base-layer until 18:00 tomorrow or 18:00 tomorrow until 18:00 on the day after production. tomorrow. Hence, IS = (0,0) corresponds to normal, uninterrupted Table 6 summarizes the snowpack models, the temporal and spatial production according to the settings defined earlier, whereas IS = (1,1) resolutions, and the values of the snow management related parameters indicates that all production should be ceased for the next two days. used for the individual simulations. Combining these settings (two scenarios each for PR, TT, and CT, and four scenarios for IS) amounts to 32 separate simulations. In ad- 5. Results and discussion dition, a model run considering untreated natural snow only and a run considering groomed natural snow without snowmaking are included In the following, simulation results for each ski resort obtained as well, amounting to a total of 34 combinations as shown in Fig. 5. using the model parameterizations described above are presented. Model runs were performed for two winter seasons representing two 4.3. Model setup comparatively extreme cases in terms of snow characteristics: 2016/17, one of the driest winters in the recent years with many regions in the In this study, we present simulation results for all nine pilot ski Alps receiving almost no snow until early January, and 2017/18, an resorts demonstrating the functionality of the snow management especially snow-rich winter throughout the entire Alps with large modules of the three snowpack models and the ability to account for snowfalls already in November and December in some regions. For different snow management configurations. The models were first set these two seasons, simulations were carried out for all ski resorts using up for each ski resort for natural snow conditions using standard con- all ten strategic snow management configurations, i. e., (numbers ac- figurations of the respective snowpack models. The AMUNDSEN and cording to Fig. 5) the natural snow-only configurations 1 (no grooming) SNOWPACK/Alpine3D simulations were performed in fully distributed and 2 (with grooming), and the snowmaking-enabled configurations 3, mode using a temporal resolution of 1 h and spatial resolutions between 7, 11, 15, 19, 23, 27, and 31, i. e., accounting for both fan guns and 5 and 15 m. The Crocus simulations on the other hand were performed lance guns as well as different temperature thresholds and base-layer directly on the SRU scale, i. e., by running independent 1-D Crocus production targets. simulations for each SRU. SRUs for the two French resorts were derived The results for each ski resort are presented as follows:

7 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Fig. 5. The 34 model conﬁgurations as deﬁned in Table 5.

• Maps showing the simulated SWE distribution for 24 December the AMUNDSEN and SNOWPACK/Alpine3D simulations are shown for 2016 and 2017, i. e., at the beginning of the economically crucial the entire model grids, i. e., including untreated natural snow beside the Christmas holiday period. The Crocus simulation results for the French slopes and without aggregation of the model outputs to the SRU scale. resorts are shown for the the deﬁned slope SRUs, whereas the results for • Time series of simulated snow depth, SWE, and density for one

Table 6 Snowpack models, temporal and spatial resolution, and values of the snow management related parameters for the simulations presented in this study.

Colfosco, San Vigilio, Seefeld La Plagne, Les Saisies Arosa-Lenzerheide Garmisch Classic/ Livigno Obergurgl Zugspitze

Snowpack model AMUNDSEN AMUNDSEN Crocus SNOWPACK/ SNOWPACK/Alpine3D SNOWPACK/ Alpine3D Alpine3D Temporal resolution 1 h 1 h 1 h 1 h 1 h 1 h Spatial resolution 10 m 5 m SRU scalea 10 m 5 m 15 m

Snowmaking parameters

PPb 01 Nov–15 Dec 01 Nov–15 Dec 01 Nov–15 Dec 01 Nov–15 Dec 15 Nov–15 Dec 01 Nov–15 Dec

PPr 16 Dec–31 Mar 16 Dec–31 Mar 16 Dec–31 Mar 16 Dec–31 Mar 16 Dec–01 Mar 16 Dec–31 Mar

PTb 00:00–24:00 00:00–24:00 00:00–24:00 00:00–24:00 00:00–24:00 00:00–24:00

PTr 18:00–08:00 18:00–08:00 18:00–08:00 17:00–08:00 17:00–08:00 17:00–08:00 − CT (kg m 2) (see Table 5)

STb (m) (see Table 5)

STr (m) 0.6 0.6 0.6 0.6 0.6 0.6 TT (°C) (see Table 5) − WT (m s 1) 4.2 4.2 4.2 2.8 2.8 2.8 NG Slope-dependent Slope-dependent n/a Slope-dependent Slope-dependent Slope-dependent SS (m2) n/a n/a 3300 n/a n/a n/a 3 −1 −1 PRa (m h C ) (see Table 5) 3 −1 PRb (m h ) (see Table 5) WA (m3) Undisclosedb Undisclosedb n/a Unlimited Undisclosed2 Unlimited − RR (m3 h 1) 0 0 n/a 0 0 0 − FT (m3 h 1) Undisclosedb Undisclosedb n/a Unlimited Unlimited Unlimited WL 0.3 0.3 0.4 0.2 0.2 0.2 −3 ρmm (kg m ) 400 400 500 Eq. (2) Eq. (2) Eq. (2) 2 −1 SSAmm (m kg ) n/a n/a 23 n/a n/a n/a

Smm (%) n/a n/a 0.9 1 1 1 Grooming parameters GP 01 Nov–30 Apr 01 Nov–30 Apr 01 Nov–30 Apr 01 Nov–30 Apr 15 Nov–31 Mar 01 Nov–30 Apr GT 21:00–22:00 21:00–22:00 20:00–21:00 (and 06:00–09:00 in 21:00–22:00 21:00–22:00 21:00–22:00 case of snowfall) − − − GH 20 kg m 2 20 kg m 2 20 kg m 2 0.4 m 0.4 m 0.4 m − PD n/a n/a 35 kg m 2 0.4 m 0.4 m 0.4 m −3 ρt (kg m ) 450 450 450 450 400 450 2 −1 SSAt (m kg ) n/a n/a 25 n/a n/a n/a

St (%) n/a n/a 0.9 n/a n/a n/a

a SRUs were derived using a 30 m DEM. Mean SRU area: 1.3 ha. b Conﬁdential information not to be disclosed.

8 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Fig. 6. AMUNDSEN simulation results for the Obergurgl ski resort. Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the slope sections covered by snowmaking configuration 23 was used, remaining slope sections were subject to grooming only, and off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 1938 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed). Observed natural snow depth from a nearby location (indicated with a red triangle in the maps) is shown in black. selected slope location within the resort based on simulations for the increased compaction. ten configurations. Additionally, both observed natural snow depth Results of the model runs for the snowmaking-enabled configura- from a representative nearby off-slope location (for evaluating simu- tions are shown in the figures in form of the mean over all configura- lation results for configuration 1) as well as on-slope snow depth tions and the range (min–max), shown as solid orange lines and shaded measured using grooming devices (for evaluating simulation results for bands, respectively. The results show that in both seasons first snow the snowmaking-enabled configurations) are included in the results production generally starts in early to mid-November in most resorts, i. (where available). e., as soon as temperatures are low enough to allow production. Each For the figures shown in this section, one representative resort was snow gun then produces snow until the specified production target (in chosen for each snowpack model: Obergurgl for AMUNDSEN (Fig. 6), the case of AMUNDSEN and Crocus) or snow depth threshold (in the Les Saisies for Crocus (Fig. 7), and Arosa-Lenzerheide for SNOWPACK/ case of SNOWPACK/Alpine3D) is reached. Especially in the 2016/17 Alpine3D (Fig. 8). Results for the remaining resorts are shown in Ap- season, due to lacking natural snowfalls in most resorts snow depth is pendix A. still below the preset threshold (60 cm) at the start of the reinforcement It should be emphasized that when interpreting the snow depth time period (16 December) for all configurations, thus production is resumed series, the differences in density of groomed vs. ungroomed snow must starting from this date as soon as the ambient conditions allow it and be taken into account. As the figures show, snow depth of ungroomed depending on if water is available. In some resorts (e. g., Garmisch locations are often comparable or even exceed snow depth of areas Classic/Zugspitze), snow is then still sporadically produced throughout covered by snowmaking, despite the latter corresponding to a sig- January in order to maintain the 60 cm threshold. In 2017/18, large nificantly higher snow mass. natural snowfalls occur especially during December and January, with As shown in the line plots, natural snow accumulation is, especially peak (natural) SWE amounts almost quadrupling compared to the in early winter, vastly different in the two considered seasons. In 2017/ previous season in some resorts. Hence, during this season snow pro- 18 for several of the resorts natural snow depth exceeds 1 m (corre- duction is completely stopped after the base-layer period in most cases. sponding to > 50cm after grooming) already in early December, while This also explains the differences in the spread of the simulation results in 2016/17 despite some early snowfalls in November the slopes are for the different management configurations (orange shaded areas in practically snow free until the end of December in some resorts, only the line plots) between the two seasons: for most resorts the spread receiving noteworthy natural snowfalls from January onwards. generated by the different configurations converges into a single line Results of configuration 2 (groomed natural snow) reflect the after the base-layer period in 2016/17, whereas in 2017/18 there is still parameter settings of the grooming modules. For shallow snowpacks in some variation between the configurations. Since in 2016/17 snow the early season, results of configuration 1 and 2 are identical as depth is still below the 60 cm threshold after the end of the base-layer grooming in the models is only performed for snowpacks reaching a period, in the reinforcement period for all configurations snow is pro- certain minimum SWE or depth (as specified by the GH parameter). As duced until exactly 60 cm are reached, thus resulting in the same soon as this threshold is exceeded the daily grooming schedule is per- snowpack evolution for all configurations for the remainder of the formed, resulting in a marked decrease in snow depth due to the season (since no more snow is produced afterwards). In 2017/18, due

9 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Fig. 7. Crocus simulation results for the Les Saisies ski resort. Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the SRUs covered by snowmaking configuration 23 was used, remaining SRUs were subject to grooming only). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 1500 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed). Observed natural snow depth from a nearby location (indicated with a red triangle in the maps) as well as observed on-slope snow depth are shown in black and yellow, respectively. to more natural snowfall the 60 cm threshold is already exceeded resorts. However, both the visual comparison in the result figures and during the base-layer period, hence the only snow production in this the computed model efficiency values in the right side of Table 7 show season is due to the fixed base-layer production volumes (150 and that in several cases even these generic model configurations lead to − 250 kg m 2 – i. e., after the base-layer period the spread shown in the very satisfactory simulation results for the selected points. line plots actually consists only of two distinct lines corresponding to Finally, in the maps showing the spatially distributed model output these values). for 24 December, the differences between the two seasons are again Comparison of the available natural snow depth observations (black clearly visible, showing large natural snow amounts in 2017 and – for lines in the figures) with the simulation results shows that the natural several resorts – virtually snow-free mountains in 2016. For the ski snow depth evolution for the given locations is reasonably well re- slopes, the areas equipped by snowmaking facilities can be clearly produced by the models. This is confirmed by the quantitative eva- distinguished from slopes covered by natural snow only. While in some luation of model performance for natural snow, shown in the left part of resorts (e. g., Lenzerheide and Livigno) only few slopes are covered by Table 7. Here, model performance in reproducing natural snow evolu- snow guns as visible in the results, several other resorts have near-full tion for the selected points is evaluated using three efficiency criteria: snowmaking coverage. Simulated SWE amounts on slopes covered by the coefficient of determination R2 (range between 0 and 1, higher snowmaking are considerably larger and more homogeneous due to the values are better), the Kling-Gupta efficiency KGE (range between −∞ set production targets or snow depth thresholds in the base-layer and 1, higher values are better), and the bias (i. e., the average tendency period, ensuring that production is stopped as soon as the respective of the simulated values to be larger (resulting in positive bias values) or snow amounts have been produced or the defined snow depth has been smaller (negative values) than the observations). The values in Table 7 reached. were calculated for the two considered seasons (Oct–May 2016/17 and 2017/18) where possible. While the temporal availability of observa- 6. Conclusions and outlook tions differs between the resorts and the values are hence not directly comparable (e. g., for La Plagne only very few observations are avail- The initiative for this study emerged within the H2020 PROSNOW able (Fig. A.12)), for all resorts a satisfactory model performance in project, where a forecasting system for snow conditions in ski resorts terms of natural snow depth is revealed. based on meteorological forecasts and snowpack modeling is being The same efficiency criteria were calculated for the on-slope loca- developed. While the snowpack models AMUNDSEN and Crocus were tions using snow depth measurements from grooming devices and the already equipped with snow management modules presented in pre- simulation results for the eight snowmaking-enabled snow management vious studies, as part of the work in PROSNOW a new snow manage- configurations. Deviations between the observed and simulated snow ment scheme was integrated into SNOWPACK/Alpine3D. These three depths do not reflect model deficiencies, but the comparison rather implementations now represent the state of the art of the integration of shows to which degree the generic set of snow management config- snow management processes in physically based snowpack models. urations as presented herein might be applicable for the individual ski While the specific implementations differ to some degree between the

10 .Hne,e al. et Hanzer, F. 11 Cold RegionsScienceandTechnology172(2020)102995

Fig. 8. SNOWPACK/Alpine3D simulation results for the Lenzerheide ski resort. Maps (displaying only parts of the resort for visualization purposes) show the simulated SWE distribution for 24 December 2016 and 2017 (for the slope sections covered by snowmaking configuration 23 was used, remaining slope sections were subject to grooming only, and off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 1745 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed). Observed on-slope snow depth is shown in yellow. F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Table 7 Model efficiency criteria (coefficient of determination R2, Kling-Gupta efficiency KGE, and bias) calculated using observed and simulated natural snow depth as well as observed and simulated on-slope snow depth (for the latter, mean as well as minimum and maximum values over the eight snowmaking-enabled configurations are listed) for the seasons 2016/17 and 2017/18. Temporal availability of observation data varies among the resorts. For Garmisch Classic/Zugspitze and Livigno, neither natural nor on-slope snow depth observations were available.

Natural snow Managed snow

R2 KGE Bias R2 KGE Bias

Arosa-Lenzerheide 0.95 (0.94–0.96) 0.42 (0.22–0.64) 0.40 (0.09–0.68) Colfosco 0.91 0.84 −0.03 0.78 (0.76–0.80) 0.58 (0.49–0.66) −0.10 (−0.18–−0.03) La Plagne 0.96 0.75 0.23 Les Saisies 0.99 0.72 −0.19 0.87 (0.84–0.90) 0.71 (0.68–0.75) 0.27 (0.22–0.31) Obergurgl 0.98 0.80 0.16 0.49 (0.48–0.49) −0.09 (−0.10–−0.07) −0.06 (−0.11–−0.00) San Vigilio 0.94 (0.93–0.94) 0.06 (0.04–0.08) −0.34 (−0.45–−0.26) Seefeld 0.95 0.75 0.21 0.87 (0.87–0.87) 0.59 (0.57–0.61) 0.15 (0.07–0.22) models, all three share a certain set of core parameters that can be used evenly over a predefined slope section, and that grooming has no effects to adapt the models to resort-specific practices. These implementations on the distribution of snow but rather only compacts it. In practice, allow for a very detailed simulation of snow management practices in snow is heavily redistributed on a daily basis both due to skier move- the three modeling systems, taking into account snow demand (being ments and due to grooming operations (involving both moving snow largely influenced by socioeconomic considerations), the meteor- from slope locations to other slope locations as well as moving natural ological conditions in terms of wet-bulb temperature and wind speed, snow from off-slope locations to the slopes). Furthermore, snowmaking and the ski resort infrastructure that ultimately determines the amount losses due to thermodynamic and mechanical effects are currently only of snow that can actually be produced in a given time step. The im- represented using a fixed water loss factor in the models. In practice, plemented grooming schemes allow to account for the distinct prop- these losses are highly variable depending on meteorological conditions erties of groomed snow on ski slopes depending on the amount of snow and other factors such as the exact positioning of the snow guns de- present and a defined grooming schedule. pending on current wind conditions (which determines if wind-blown To provide ski resort representatives with a range of possible future snow is transported to off-slope locations and is actually “lost”,orifitis snowpack states depending on their management choices, a set of de- transported to other parts of the slopes or off-slope locations where it fault model configurations has been prepared that can be further re- can be easily recovered). Third, the assumption that snow guns are fined for each ski resort and subsequently be used for the operational assigned to fixed locations is often not true in reality, where frequently simulations. at least part of the snow guns are mobile and moved around during the For the results presented in this study, the models were set up for all season, depending on where snow is currently required. nine pilot ski resorts. Simulation runs were performed for two con- Since many of these processes currently cannot be reasonably re- trasting seasons: the very dry 2016/17 season and the snow-rich 2017/ presented in a numerical model, deviations between the real snow 18 season. All ten strategic snow management configurations (un- conditions on the slopes and those simulated by the models are ungroomed natural snow, groomed natural snow, and eight snowmaking- avoidable. Hence, for operational model applications such as envisaged enabled configurations) were used for the presented results, allowing to in the PROSNOW project, it is crucial to integrate local observations demonstrate the models’ capabilities to reproduce the observed natural (water consumption recordings from the snow guns and distributed snow evolution, and to show the functionality of the snowmaking and snow depth measurements from groomers) in order to be able to pro- grooming modules based on the selected parameters. duce reliable forecasts. Future work will focus on updating model states It should be emphasized that the results of the snow production runs by assimilating these local observations, as well as evaluating the do not reflect the actual snow management practice for these seasons in snowpack simulations driven by meteorological forecasts. First eva- the individual ski resorts, but are rather solely intended to demonstrate luations of Crocus snowpack simulations (natural snow only) driven by the functionality of the snow management modules. The same snow meteorological forecasts in the context of PROSNOW are presented in management configurations were chosen for all ski resorts in order to Carmagnola et al. (2018). The next steps of the PROSNOW project, ensure comparability of the model results between the different resorts. which are beyond the scope of this paper and will be explored in future In reality, management practices between the resorts might vary works, will first consist of using forecast products to drive the snow widely, and both the fixed model parameters and the set of model simulations with an ensemble of meteorological members, in order to configurations will have to be refined accordingly for each resort. anticipate the future conditions of the snowpack on the ski slopes, However, the presented approaches for simulating snow manage- which is the main interest of the PROSNOW project. In addition, local ment also have some limitations, which become partly visible in the measurements of water consumption for snowmaking and snow depth presented comparison of the observed and simulated snow depths on on the ski slopes will be inserted in the modeling chain, to improve the the slopes. On the one hand, obviously the presented assumptions for description of the initial state of the snowpack before running the the decision when and where snow should be produced and when the forecast simulations. Subsequently, the performances of the models will slopes should be groomed are highly simplified and cannot be expected be evaluated by comparing model outputs with local and remotely- to be an accurate depiction of reality. The presented approach to run sensed measurements, not only in terms of snow depth and SWE, but the models with different management configurations is one attempt to also looking at other variables such as the amount of water consumed account for these simplifications and present the end users of the for snowmaking over the season. modeling chain with a range of possible snowpack evolutions. The ac- Finally, while in this paper each ski resort was simulated only with curacy of the model results is in this case highly dependent on the in- one single snowpack model, from the model development perspective volvement of the stakeholders, i. e., to which degree the real-world an intercomparison of the three snowpack models being applied to the snow management practice can be translated to the model im- same resort(s) would be a logical next step. Differences in the simula- plementations (Strasser et al., 2014). On the snowpack modeling side, tion results of the three models for a given location would be mainly one major assumption in all three model implementations is that for a due to different approaches in the simulation of snow management on given snow gun all of the produced snow is distributed immediately and the one hand and the different approaches for simulating the snowpack

12 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995 energy and mass balance on the other hand. For a reasonable com- Editing parison these eﬀects would have to be disentangled and discussed ac- Franziska Koch: Software, Resources, Writing - Reviewing and cordingly, which is not straightforward and out of scope for this paper Editing but which we will consider for the future. Rather, here we have shown Fabiano Monti: Resources that all three models are able to produce plausible and robust results on Mathias Bavay: Software the ski slope scale, and that the accuracy of the results is mainly de- Matthias Bernhardt: Supervision pendent on the degree to which the real-world snow management Matthieu Lafaysse: Software practices are integrated. Michael Lehning: Supervision, Writing - Reviewing and Editing Ulrich Strasser: Conceptualization, Methodology, Supervision, Data availability Writing - Reviewing and Editing Hugues François: Software, Resources Datasets related to this article can be found at https://doi.org/10. Samuel Morin: Conceptualization, Methodology, Supervision, 5281/zenodo.3564236 (Hanzer et al., 2019). Writing - Reviewing and Editing, Funding acquisition

Declaration of Competing Interests Acknowledgements The authors declare that they have no known competing ﬁnancial interests or personal relationships that could have appeared to inﬂu- We thank our project partners and pilot ski resorts for many con- ence the work reported in this paper. structive discussions which are contributing to shape model implementations that are practical from a user perspective. Furthermore Author Contribution we wish to thank Daniel Günther (University of Innsbruck) who provided the setup of the AMUNDSEN model runs used for this study, and Florian Hanzer: Conceptualization, Methodology, Software, two anonymous reviewers who provided helpful and constructive Resources, Validation, Visualization, Writing - Original Draft comments that helped to improve the manuscript. Carlo Maria Carmagnola: Conceptualization, Methodology, This project has received funding from the European Union’s Software, Resources, Writing - Reviewing and Editing Horizon 2020 research and innovation programme under grant agree- Pirmin Philipp Ebner: Software, Resources, Writing - Reviewing and ment No 730203.

Appendix A. Further results

Fig. A.9. AMUNDSEN simulation results for the Colfosco ski resort. Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the slope sections covered by snowmaking configuration 23 was used, remaining slope sections were subject to grooming only, and off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 1820 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed). Observed natural snow depth from a nearby location (indicated with a red triangle in the maps) as well as observed on-slope snow depth are shown in black and yellow, respectively.

13 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Fig. A.10. AMUNDSEN simulation results for the San Vigilio ski resort. Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the slope sections covered by snowmaking configuration 23 was used, remaining slope sections were subject to grooming only, and off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 1748 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed). Observed on-slope snow depth is shown in yellow.

Fig. A.11. AMUNDSEN simulation results for the Seefeld cross-country ski resort (2019 Nordic World Ski Championships tracks only). Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the ski tracks (all covered by snowmaking) configuration 23 was used, off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a track location at 1182 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations

14 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995 the mean (orange line) and range (min–max; shaded bands) are displayed). Observed natural snow depth from a nearby location (indicated with a red triangle in the maps) as well as observed on-slope snow depth are shown in black and yellow, respectively.

Fig. A.12. Crocus simulation results for the La Plagne ski resort. Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the SRUs covered by snowmaking configuration 23 was used, remaining SRUs were subject to configuration 2, i. e., grooming only). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 2100 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed). Observed natural snow depth from a nearby location (indicated with a red triangle in the maps) is shown in black.

Fig. A.13. SNOWPACK/Alpine3D simulation results for the Garmisch Classic/Zugspitze ski resort. Maps show the simulated SWE distribution for 24 December 2016 and 2017 (for the slope sections covered by snowmaking configuration 23 was used, remaining slope sections were subject to grooming only, and off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 1438 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking- enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed).

15 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

Fig. A.14. SNOWPACK/Alpine3D simulation results for the Livigno ski resort. Maps (displaying only parts of the resort for visualization purposes) show the simulated SWE distribution for 24 December 2016 and 2017 (for the slope sections covered by snowmaking configuration 23 was used, remaining slope sections were subject to grooming only, and off-slope areas show natural snow simulation results). Line plots show snow depth (top), SWE (center), and density (bottom) evolution for a slope location at 2540 m a.s.l. (indicated with a red circle in the maps) for the seasons 2016/17 and 2017/18 and the ten strategic snow management configurations according to Fig. 5 (for the eight snowmaking-enabled configurations the mean (orange line) and range (min–max; shaded bands) are displayed).

References snow hydrology. Hydrol. Process. 20, 2111–2128. https://doi.org/10.1002/hyp. 6204. Marke, T., Strasser, U., Hanzer, F., Stötter, J., Wilcke, R.A.I., Gobiet, A., 2015. Scenarios Abegg, B., Morin, S., Demiroglu, O.C., François, H., Rothleitner, M., Strasser, U., 2020. of future snow conditions in Styria (Austrian Alps). J. Hydrometeorol. 16, 261–277. Overloaded! Critical revision and a new conceptual approach for snow indicators in https://doi.org/10.1175/JHM-D-14-0035.1. ski tourism. Int. J. Biometeorol. https://doi.org/10.1007/s00484-019-01841-8. In Morin, S., Dubois, G., the PROSNOW Consortium, 2018. PROSNOW – Provision of a press. prediction system allowing for management and optimization of snow in Alpine ski Bartelt, P., Lehning, M., 2002. A physical SNOWPACK model for the Swiss avalanche resorts. In: ISSW 2018 Innsbruck Proceedings. warning. Cold Regions Sci. Technol. 35, 123–145. https://doi.org/10.1016/S0165- Morin, S., Horton, S., Techel, F., Bavay, M., Coléou, C., Fierz, C., Gobiet, A., Hagenmuller, 232X(02)00074-5. P., Lafaysse, M., Ližar, M., Mitterer, C., Monti, F., Müller, K., Olefs, M., Snook, J.S., Bavay, M., Egger, T., 2014. MeteolO 2.4.2: A preprocessing library for meteorological van Herwijnen, A., Vionnet, V., 2019. Application of physical snowpack models in data. Geosci. Model Dev. 7, 3135–3151. https://doi.org/10.5194/gmd-7-3135-2014. support of operational avalanche hazard forecasting: A status report on current im- Carmagnola, C.M., Morin, S., Lafaysse, M., Vernay, M., François, H., Eckert, N., Batté, L., plementations and prospects for the future. Cold Regions Sci. Technol. 102910. Soubeyroux, J.M., Viel, C., 2018. Combination of climatological information and https://doi.org/10.1016/j.coldregions.2019.102910. meteorological forecast for seamless prediction of Alpine snow conditions. In: ISSW Spandre, P., François, H., Morin, S., George-Marcelpoil, E., 2015. Snowmaking in the 2018 Innsbruck Proceedings. French Alps. Climatic context, existing facilities and outlook. J. Alp. Res.—Revue de Corripio, J., 2002. Modelling the Energy Balance of High Altitude Glacierised Basins in géographie alpine. https://doi.org/10.4000/rga.2913. the Central Andes. PhD thesis. University of Edinburgh. Spandre, P., François, H., George-Marcelpoil, E., Morin, S., 2016a. Panel based assessment Durand, Y., Brun, E., Merindol, L., Guyomarc’h, G., Lesaffre, B., Martin, E., 1993. A of snow management operations in French ski resorts. J. Outdoor Recreat. Tour. 16, meteorological estimation of relevant parameters for snow models. Ann. Glaciol. 18, 24–36. https://doi.org/10.1016/j.jort.2016.09.002. 65–71. https://doi.org/10.3189/S0260305500011277. Spandre, P., Morin, S., Lafaysse, M., Lejeune, Y., François, H., George- Marcelpoil, E., Flügel, W.A., 1995. Delineating hydrological response units by geographical information 2016b. Integration of snow management processes into a detailed snowpack model. system analyses for regional hydrological modelling using PRMS/MMS in the drai- Cold Regions Sci. Technol. 125, 48–64. https://doi.org/10.1016/j.coldregions.2016. nage basin of the River Bröl, Germany. Hydrol. Process. 9, 423–436. https://doi.org/ 01.002. 10.1002/hyp.3360090313. Spandre, P., François, H., Thibert, E., Morin, S., George-Marcelpoil, E., 2017. Grünewald, T., Wolfsperger, F., 2019. Water losses during technical snow production: Determination of snowmaking efficiency on a ski slope from observations and Results from field experiments. Front. Earth Sci. 7. https://doi.org/10.3389/feart. modelling of snowmaking events and seasonal snow accumulation. The Cryosphere 2019.00078. 11, 891–909. https://doi.org/10.5194/tc-11-891-2017. Hanzer, F., Marke, T., Strasser, U., 2014. Distributed, explicit modeling of technical snow Spandre, P., François, H., Verfaillie, D., Pons, M., Vernay, M., Lafaysse, M., George, E., production for a ski area in the Schladming region (Austrian Alps). Cold Regions Sci. Morin, S., 2018. Winter tourism and climate change in the Pyrenees and the French Technol. 108, 113–124. https://doi.org/10.1016/j.coldregions.2014.08.003. Alps: Relevance of snowmaking as a technical adaptation. The Cryosphere Disc. 1–36. Hanzer, F., Helfricht, K., Marke, T., Strasser, U., 2016. Multilevel spatiotemporal vali- https://doi.org/10.5194/tc-2018-253. dation of snow/ice mass balance and runoff modeling in glacierized catchments. The Steiger, R., Abegg, B., 2013. The sensitivity of Austrian ski areas to climate change. Tour. Cryosphere 10, 1859–1881. https://doi.org/10.5194/tc-10-1859-2016. Plan. Dev. 10, 480–493. https://doi.org/10.1080/21568316.2013.804431. Hanzer, F., Carmagnola, C.M., Ebner, P.P., Koch, F., François, H., 2019. Datasets for the Steiger, R., Mayer, M., 2008. Snowmaking and climate change. Mount. Res. Dev. 28, publication “Simulation of snow management in Alpine ski resorts using three dif- 292–298. ferent snow models”. https://doi.org/10.5281/zenodo.3564236. Steiger, R., Scott, D., Abegg, B., Pons, M., Aall, C., 2017. A critical review of climate Lafaysse, M., Cluzet, B., Dumont, M., Lejeune, Y., Vionnet, V., Morin, S., 2017. A mul- change risk for ski tourism. Curr. Issues Tour. 1–37. https://doi.org/10.1080/ tiphysical ensemble system of numerical snow modelling. The Cryosphere 11, 13683500.2017.1410110. 1173–1198. https://doi.org/10.5194/tc-11-1173-2017. Strasser, U., 2008. Modelling of the Mountain Snow Cover in the Berchtesgaden National Lehning, M., Völksch, I., Gustafsson, D., Nguyen, T.A., Stähli, M., Zappa, M., 2006. Park. Technical Report Berchtesgaden National Park, Berchtesgaden. ALPINE3D: A detailed model of mountain surface processes and its application to Strasser, U., Warscher, M., Liston, G.E., 2011. Modeling snow-canopy processes on an

16 F. Hanzer, et al. Cold Regions Science and Technology 172 (2020) 102995

idealized mountain. J. Hydrometeorol. 12, 663–677. envsoft.2014.06.014. Strasser, U., Vilsmaier, U., Prettenthaler, F., Marke, T., Steiger, R., Damm, A., Hanzer, F., Vionnet, V., Brun, E., Morin, S., Boone, A., Faroux, S., Moigne, P.L., Martin, E., Willemet, Wilcke, R.A.I., Stötter, J., 2014. Coupled component modelling for inter- and trans- J.M., 2012. The detailed snowpack scheme Crocus and its implementation in SURFEX disciplinary climate change impact research: Dimensions of integration and examples v7.2. Geosci. Model Dev. 5, 773–791. https://doi.org/10.5194/gmd-5-773-2012. of interface design. Environ. Model. Softw. 60, 180–187. https://doi.org/10.1016/j.